CN113469034A - Fusion recognition method for coal rock cutting state of coal mining machine - Google Patents

Abstract

The invention discloses a mixed optimization algorithm of MFO-FOA, and some simulations are carried out to verify the effectiveness and superiority of the algorithm. Then, the proposed MFO-FOA algorithm is utilized to find out the optimal network parameters of the RBFNN, so as to realize excellent generalization capability and prediction performance. Further, the collected signal is decomposed by variational mode decomposition, and the characteristics of the first three eigenmode function components are extracted using the envelope entropy and the kurtosis. Feature vectors obtained from the three types of sensor data are used to construct the RBFNN classifier. In addition, a D-S evidence theory with evidence correlation coefficients is introduced, and the preliminary identification results of the three RBFNN classifiers are fused. Finally, a self-designed experiment platform for cutting coal rocks by a coal mining machine is established, and a plurality of experiments are provided. Experimental results based on measured data show that the method can effectively identify the coal rock cutting state and is high in precision.

Description

Fusion recognition method for coal rock cutting state of coal mining machine

Technical Field

The invention relates to the technical field of coal mining machines, in particular to a fusion identification method for a coal rock cutting state of a coal mining machine.

Background

Coal is the most abundant and widely distributed fossil fuel on earth. Among fossil energy sources that have been explored in china, coal accounts for approximately 94%. This phenomenon has led to the fact that the energy structure of china mainly based on coal is not changed for a long time. Due to the harsh underground working environment and the low level of reliability and automation of coal mining equipment, coal production remains a high risk industry. In 2018, 224 coal mine safety accidents occurred in China. The 100 million ton mortality is 3.1 times that of the united states and 6.6 times that of australia. Therefore, in order to realize safe and efficient production of coal mines, the intelligent level of mining equipment is urgently needed to be improved. Coal mining machines are important components of coal mining equipment and are mainly responsible for cutting and transporting coal. Accurate identification of the cutting state of the coal rock of the coal mining machine is a prerequisite for realizing automatic mining, and is becoming a research hotspot in the field of coal resources.

In recent years, in order to improve the intelligent operation level of a coal mining machine, scholars at home and abroad mainly focus on two aspects of coal rock interface identification and memory cutting, and a plurality of effective methods are provided. Including the use of radar technology for identifying coal-rock interfaces, which has been industrially tested in quedrek coal mines. In pennsylvania, a tactile sensor was developed for detecting different types of material layers, in which a shearer may operate on a longwall of an underground coal mine. And a low-activity gamma ray logging technology is adopted as a sensitive tool for describing a coal-rock interface. An ultrasonic phased array coal-rock interface identification method adopting acoustic impedance difference between coal and rock and a phased array technology. The vibration signal and the infrared thermal image of the coal mining machine are deeply analyzed, so that the effective dynamic identification of coal and rock in the coal mining process of the coal mining machine is realized. In addition, many other methods have been proposed to identify coal-rock interfaces based on sound waves, ground penetrating radar, natural gamma rays and terahertz time-domain spectroscopy, and good results have been obtained in the laboratory. Through the research, the coal rock interface identification method can only identify two cutting states: coal or rock. When the geological conditions of the coal bed are changed violently, the defects of low identification precision, poor applicability and the like exist. Memory cutting is the most widely used automatic control method for coal mining machines. An improved method combining an improved genetic algorithm and a fuzzy logic control method is adopted to reduce the expansion of coal deformation in the memory cutting process of the coal mining machine. By adopting the data correlation of adjacent coal seams, a hidden Markov model memory cutting method is developed for the coal mining machine, and the adjustment frequency and the adjustment precision can be improved. Although the memory cutting method can improve the automatic control level of the coal mining machine to a certain extent, the application effect when the coal seam is suddenly broken is not ideal. In this context, some scholars attempt to determine the cutting state of the coal mining machine, providing a basis for the cutting state. The prior art has also investigated the previously overlooked impact of rock parameters and engineered rock properties on specific cutting energies, and the results can be used to initially determine the operating state of a shearer. Development of a coal bed rock fissionability grading system based on specific energy. Theoretical models of cutting load and current are constructed, and an angle analysis method of the cutting load characteristic of the coal mining machine based on particles is provided to determine the cutting state of the coal mining machine. With the vibration signal, the sound uses the signal of the cutting zone and the temperature of the cutting zone to identify the cutting state of the shearer coal rock by using some intelligent classifier. Due to the complex structure of the coal mining machine, the severe working conditions and the unreliable data of the single type of sensor, the coal-stone state of the coal mining machine cannot be accurately reflected. Therefore, the application provides a multi-sensor information fusion method for performing coal rock cutting state identification by comprehensively utilizing vibration signals and sound signals.

In the field of state recognition, there are many intelligent recognition methods, which have been widely used for fault diagnosis and image classification. Among the methods, a basic function neural network (RBFNN) is proposed by Moody and Darken in 1988, and has the advantages of simple network structure, strong approximation capability, high learning speed, difficulty in falling into local minimum problem and good robustness. In order to improve the learning performance of RBFNNs, many researchers have adopted RBFNN structures and some meta-heuristic algorithms such as Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO) and Genetic Algorithm (GA) which are naturally heuristic to realize network learning. However, these techniques always have some drawbacks. The intelligent optimization algorithm (FOA), as a novel group intelligent search algorithm, is widely applied to many fields due to its simple structure, strong global optimization capability, and easy understanding and learning. Like other optimization algorithms, FOAs may also present problems such as premature maturation and poor prospecting capabilities. Hybridization is a rational way to gain advantage and avoid disadvantages. Considering that the moth optimization (MFO) algorithm has a unique population update mechanism, the present study aims to design a hybrid algorithm of PSO and MFO called MFO-FOA for training RBFNN and finding the best network parameters for proper performance verification and validation. However, each technique always has some drawbacks. The single type of sensor data is not reliable enough under actual working conditions and may reduce the recognition effect of the coal cutter on the cutting state of the coal rock based on the proposed classifier based on the RBFNN model.

Disclosure of Invention

In order to solve the technical problems, the invention provides a fusion identification method of coal cutter coal rock cutting states, and a D-S evidence theory is introduced to realize decision-level fusion of identification results based on a single signal source.

The technical purpose of the invention is realized by the following technical scheme:

a fusion recognition method for coal cutter coal rock cutting states comprises the following steps:

s1: in the coal rock cutting process, data collection is carried out through a sound sensor and a vibration sensor;

s2: identifying the collected data through an RBF classifier;

s3: and fusing the identification results of the RBF classifier by using a D-S evidence theory to obtain a final identification result.

Preferably, the S1 process includes a sound sensor and two vibration sensors respectively.

As a preferred solution, in S2, the RBF classifier includes an input layer, a hidden layer and an output layer, wherein the function of the RBF is expressed as:

wherein, | | x_p-c_i||²Is the norm of the euclidean norm,

is the p-th input sample, c_iIs the center of the gaussian function of the ith node, and σ is the variance, i.e., the width of the gaussian function;

the output calculation method of the RBF classifier comprises the following steps:

wherein ω is_ijIs the connection weight between the hidden layer and the output layer, h is the number of nodes in the hidden layer, θ_jIs the threshold for the jth output node, which is the actual output value of j-, the first output node in the output layer.

As a preferred scheme, in the process of S2, an MFO-FOA hybrid optimization algorithm is used in the RBF classifier for classification, the random flight pattern of the fruit flies is changed to a spiral flight path, the center of the spiral flight path is the current position of the fruit fly population, and the specific algorithm is as follows:

where dis _ X_iAnd dis _ Y_iRespectively, a single position X of the previous generation_{i_last}And Y_{i_last}With current group position (X)_{i_last}And Y_{i_last}) X _ axis and Y _ axis are the initial positions of the drosophila population;

the specific implementation steps are as follows:

t1: setting a population size N, a maximum iteration number T, a position range and a random close range FR of a fruit fly population LR; randomly generating an initial Drosophila population position (X _ axis, Y _ axis); the position of the individual fruit flies is then updated:

t2: calculating the distance between the current position and the origin to obtain an odor concentration judgment value (S)_i)

T3: judging the odor concentration (S)_i) Input to (Smell)_i) In the judgment function, calculating the odor concentration Smelli of the y position of a single fruit; the best odor concentration was obtained by ranking all Smelli, and then determining and finding the corresponding single position bestIndex:

t4: comparing the current bestmdex with the global optimum concentration Smellbest; if bestIndex outperforms Smellbest, the global optimum concentration and corresponding location are updated:

t5: individual positions of fruit flies were updated:

t6: if the termination condition is met, stopping iteration; otherwise, steps T2-T5 are repeated until the maximum number of iterations T is reached.

As a preferred scheme, in the process of S3, evidence correlation coefficient measurement is introduced, a weight mean combination model is used for performing coefficient operation between the evidences, and the original evidences are modified or preprocessed;

for two evidences m in the recognition framework Θ_iAnd m_jThe correlation coefficient r is calculated by the following formula_BPAAnd degree of correlation c:

m is to be_iAnd m_jSupport degree of correlation coefficient between them as evidence, Sup (m)_i，m_i)＝r_BPA(m_i，m_j) Simply referred to as S_ijObtaining a support matrix of the evidence:

other evidence is evidence m_iThe total support of (1) is:

the normalized support can be used as evidence m_iReliability of (c) Crd (m)_i) The calculation formula is as follows:

confidence level Crd (m)_i) As a weight given to the evidence, a new evidence is obtained by averaging all evidences according to the weight, achieving fusion of highly conflicting evidences.

In conclusion, the invention has the following beneficial effects:

(1) based on the advantages and disadvantages of FOA and MFO, the FOA is improved by using a position updating mechanism of MFO, and an MFO-FOA mixed group intelligent optimization algorithm is provided. The simulation result verifies the effectiveness and superiority of the MFO-FOA algorithm.

(2) And decomposing the sound signal and the vibration signal by using the VMD, and realizing feature extraction by using the envelope entropy and the kurtosis. And finding out the optimal network parameters of the RBFNN by using the proposed MFO-FOA method, and identifying the cutting state of the coal rock according to the characteristic information of the single signal source.

(3) And constructing a fusion model based on a D-S evidence theory based on a comprehensive decision idea to make up the inconsistency of the single recognition result. Corresponding experiments show that the recognition accuracy of the fusion model can reach 98 percent and is obviously higher than that of a single signal characteristic. The performance of the method was compared under the same experimental sample. It can be concluded that the method has a certain improvement in pattern recognition, and that the superior performance of the method is demonstrated by the higher recognition accuracy.

Drawings

Fig. 1 is a block diagram of an RBF neural network in an embodiment of the present invention.

Detailed Description

This specification and claims do not intend to distinguish between components that differ in name but not function. In the following description and in the claims, the terms "include" and "comprise" are used in an open-ended fashion, and thus should be interpreted to mean "include, but not limited to. "substantially" means within an acceptable error range, within which a person skilled in the art can solve the technical problem to substantially achieve the technical result.

The terms in the upper, lower, left, right and the like in the specification and the claims are used for further explanation, so that the application is more convenient to understand and is not limited to the application.

The present invention will be described in further detail with reference to the accompanying drawings.

The main contributions of the proposed recognition scheme can be summarized as follows: (1) an optimization algorithm based on combination of MFO and FOA is provided to solve the optimal network parameters of RBFNN. The validity and superiority of the MFO-FOA are verified by comparing simulation with other popular metaheuristic algorithms. (2) A new identification technology based on RBFNN is provided, the algorithm is optimized by using MFO-FOA, and the coal and stone state identification based on three types of perception is realized. Data was collected from one sound sensor and two vibration sensors. And D-S evidence theory is applied to fuse the identification results of the three RBFNN classifiers, so that an accurate identification result is obtained. (3) Several experimental comparative studies were performed to demonstrate the effectiveness and superiority of the proposed method. The rest of the study is arranged as follows. The basic algorithm and DS evidence theory of RBF neural network are briefly introduced. In the third section, a hybrid optimization algorithm is proposed and the MFO-FOA based RBFNN parameter optimization program is introduced in detail. A proposed fusion recognition system was constructed. Some experiments were performed to verify the performance of the proposed method and compared with other methods to demonstrate its superior performance in terms of identification accuracy. Finally, the conclusion of this work is summarized.

RBF neural network

The RBF neural network is a three-layer feed-forward neural network, generally consisting of an input layer, a hidden layer and an output layer, as shown in fig. 1.

The radial basis function commonly used in RBF neural networks is a gaussian function and can be expressed as:

wherein | | | x_p-c_i||²Is the Euclidean norm

Is the p-th input sample, c_iIs the center of the gaussian function of the ith node and σ is the variance, i.e., the width of the gaussian function. The output of the RBF neural network can be calculated as:

wherein ω is_ijIs the connection weight between the hidden layer and the output layer, h is the number of nodes in the hidden layer, θ_jIs the threshold for the jth output node, which is the first output node in the output layer of the actual output value of j-.

Theory of D-S evidence

In D-S evidence theory, if a finite complete set of elements Θ ═ θ₁，θ₂，…，θ_NB, }; if g is mutually exclusive, the set may be referred to as an identification framework.

Defining a function m: 2^Θ[0，1]，

m (A) is called Basic Probability Assignment (BPA).

The confidence of proposition a can be measured by two concepts: bel (A) and Pt (A).

Bel (A) represents the overall confidence level, which can be defined as follows:

therefore, bel (a) ═ m (a) and the confidence function satisfy the following condition:

Bel(Θ)＝1

pl (A) represents the rationality belief level and can be defined as follows:

in D-S evidence theory, the Dempster rule is used to merge information from multiple independent sources. Suppose m₁And m₂BPAs, the corresponding focus element is A₁,A₂，...，A_kAnd B₁，B₂，...，B_kM represents m₁And m₂Combined new evidence. The Dempster combination rule may be described as follows:

where k represents the coefficient of collision between the evidences, which can be expressed as:

evidence correlation coefficient

Clearly, the combination rule of D-S evidence theory has inevitable drawbacks. That is, once there is some conflict between the evidences, the fusion result runs counter or inconsistent with some of the evidences. In order to enhance the fusion effect and improve the applicability, evidence correlation coefficients are introduced for measurement, a weight mean combination model is used for performing coefficient operation between the evidences, and the original evidences are modified or preprocessed.

For two evidences m in the recognition framework Θ_iAnd m_jThe correlation coefficient (expressed as r) can be calculated by the following formula_BPA) And degree of correlation (denoted c):

if the correlation coefficient between some evidence and other evidence is large, the more the evidence can be supported, the higher the confidence of the evidence and therefore the greater the weight assigned to it. The weight assigned to each evidence is measured against the confidence level of the evidence and the basic probability assignment of the evidence is processed using a weighted mean method. By the thought, m can be adjusted_iAnd m_jSupport degree of correlation coefficient between them as evidence, Sup (m)_i，m_im＝r_BPA(m_i，m_j) Hereinafter abbreviated as S_ij. Thus, a support matrix of evidence can be obtained.

Obviously, the smaller the contradiction between the two evidences, the larger the correlation coefficient and the mutual support. Other evidence is evidence m_iThe total support of (c) is:

the normalized support can be used as confidence Crd (m) of the evidence mi_i) The calculation formula is as follows:

confidence level Crd (m)_i) Can be used as a weight given to the evidence and can be obtained by averaging all evidences according to the weightNew evidence, and thus a fusion of highly conflicting evidence can be achieved.

Improved RBF neural network based on hybrid optimization algorithm

Fruit fly optimization algorithm

Pan proposed the Fruit _ y optimization algorithm (FOA) in 2012. Similar to other meta-heuristic intelligent algorithms, FOA derives from the foraging process of the natural population. During foraging, fruit flies continually exchange food information to gain the most efficient route.

Moth optimization algorithm

The moth optimization (MFO) algorithm proposed by miljalii (mirjalii) in 2015 is a novel swarm intelligence optimization algorithm. The main inspiration of the algorithm is the method of navigation of the moth in nature, i.e. lateral orientation. In the d-dimensional search space, there is a population M consisting of n moths. Each moth has a unique _ ame corresponding to it. The matrix F consisting of all _ ames is the same dimension as the moth population M. Moth is the actual search object that moves in the search space, while _ ame is the best position for moth-type searches to date. Thus, if a better solution is found, each moth will search nearby and update. By this mechanism they will never miss the best solution.

Mixed optimization algorithm based on MFO and FOA

In the FOA, the position update of each fruit fly only depends on the position of the current fruit fly population, so that the algorithm has better global search capability, but weakens local search capability and is easy to fall into local optimum. The MFO algorithm has a special location update mechanism, so the search space is not limited to the space between the moth and the flame, but also includes the whole space around the flame, thereby expanding the search space of the algorithm. Based on the advantages and disadvantages of FOA and MFO, the FOA is improved by using a position updating mechanism of MFO, and a new group intelligent optimization algorithm called MFO-FOA is formed. The specific improved idea can be described as follows:

in the process of updating the position of the fruit fly, the random flight pattern of the fruit fly is changed into a spiral flight path. The center of the spiral flight path is the current position of the fruit fly population, which increases the flight distance and enlarges the search space, so that the algorithm is not easy to fall into local optimization. Meanwhile, with the increase of the iteration times, the shape constant of the spiral equation can be gradually reduced, so that the search space of the drosophila is reduced, and the rapid convergence of the algorithm in subsequent iterations is ensured. The following were used:

where neutralization dis _ X_i and dis_Y_iRespectively, a single position X of the previous generation_{i_last} and Y_{i_last}With current group position (X)_{i_last} and Y_{i_last}) The distance between them. The specific implementation steps of b ═ 1-n/(T +1), T ═ 2-n/T) x rand +1 are as follows:

step 1: the population size N, the maximum iteration number T, the position range of the fruit fly population LR and the random close range FR are set. The initial Drosophila population positions (X _ axis, Y _ axis) were randomly generated.

The location of the individual fruit flies is then updated.

Step 2: calculating the distance between the current position and the origin to obtain an odor concentration judgment value (S)_i)。

And step 3: then, the value (S) is judged by the odor concentration_i) Input to (Smell)_i) In the decision function (also called tness function), the odor concentration Smelli at the y-position of a single fruit is calculated. By ranking all Smelli, the best odor can be obtainedConcentration, and then the corresponding single position bestIndex is determined and found.

And 4, step 4: the current bestmndex is compared to the global optimum concentration of smelbest. If bestIndexl is better than Smellbest, the global optimum concentration and corresponding location should be updated.

And 5: updating single position of fruit fly

Step 6: if the termination condition is met, the iteration is stopped. Otherwise, repeating the steps 2-5 until the maximum iteration number T is reached.

To verify the effectiveness and superiority of the MFO-FOA algorithm, the present application uses seven widely used test functions, whose plots typically have some peaks and valleys and channels. Thus, the optimization performance of the proposed algorithm and the ability to avoid falling into local optima have undoubtedly been demonstrated.

Table 1 lists the functional expressions and the theoretical optimal solution.

TABLE 1

In this simulation, four other popular intelligent algorithms, including FOA, were compared MFO, Bat Algorithm (BA) and PSO with the algorithms proposed in this application to show their distinctiveness and superiority. To verify the performance of the FOA improvement, the improved FOA (ifoa) proposed in "mechanical acoustic wavelet threshold denoising method (2016) based on improved drosophila optimization algorithm" was also used in the simulation. The main parameters are set as follows: overall size ND30, maximum iteration number TD 200. Other parameters are set to default values. To ensure accuracy, each algorithm was performed 20 times on a single function, the best obtained after 20 simulations and the average best of 20 replicates were used to measure the performance of 6 algorithms. Table 2 gives the optimization results for seven functions:

TABLE 2

As can be seen from Table 2, the optimal value of MFO-FOA is significantly higher in accuracy level than the other _ ve algorithm. The accuracy rating of MFO-FOA may be up to E-38, while the approximate accuracy rating is E-26 for MFO. The accuracy level of the other four algorithms is only E-3. The comparison result shows that the algorithm has better search performance and exploration capability in the subsequent iteration period. Particularly for the CrownedCross function, the optimal values based on MFO-FOA and MFO can reach 1.0000E-04 and 1.0036E-04, which are substantially equal to the theoretical optimal solution. It has also been shown that the convergence speed of MFO-FOA is faster than that of MFO algorithm. In addition, the location of the optimal solution for other algorithms is clearly far from the theoretical optimal point. All results demonstrate faster search speed and better convergence accuracy of MFO-FOA. The method adopts a new position updating strategy, so that the fruit flies are uniformly distributed, the global optimization is benefited, and the premature problem of FOA, BA and PSO algorithms can be effectively solved.

Comparison with other methods

In order to analyze the differences in recognition accuracy between the proposed method and other methods of our research group, it was compared to existing analyses. And (4) fusion recognition results of coal cutter coal rock cutting states based on the data of the three types of sensors. The vibration signal is taken as an analysis object and is decomposed through local mean decomposition. And extracting time-frequency characteristics, and classifying the cutting state of the coal mining machine through a fuzzy C-mean clustering algorithm. In the shearer cutting state recognition and fuzzy mean clustering, which is based on vibration-based signal analysis local mean decomposition, the cutting sound signal is taken as an analysis object and is decomposed by the modified EEMD. After the features are obtained, a probabilistic neural network is used as a classifier. In the on-line recognition of the cutting state of the shearer using the thermal infrared imager of the cutting unit, the temperature of the cutting region is used as an analysis object, and the cutting state is recognized by a Support Vector Machine (SVM). The experimental protocol was the same as the above simulation. Although excellent signal processing and feature extraction algorithms are used in the cutting state of the thermal infrared imager using the cutting unit for the identification of the cutting state of the coal cutter and the fuzzy mean clustering and the on-line identification of the coal cutter in the signal analysis of the vibration, there is also a certain degree of misjudgment, which is mainly caused by insufficient feature information in the data of a single type of sensor. In addition, due to the hysteresis of temperature transfer, the temperature of the coal mining area may not reflect different coal mining states in time. In conclusion, the fusion identification method based on RBFNN and D-S evidence theory can obtain better classification results and is superior to other competition methods. For information conflict or error information caused by sensor failure, the fusion system can solve the fault tolerance problem and make appropriate response.

The present embodiment is only for explaining the present invention, and it is not limited to the present invention, and those skilled in the art can make modifications of the present embodiment without inventive contribution as needed after reading the present specification, but all of them are protected by patent law within the scope of the claims of the present invention.

Claims (5)

1. A fusion recognition method for coal cutter coal rock cutting states is characterized by comprising the following steps:

s1: in the coal rock cutting process, data collection is carried out through a sound sensor and a vibration sensor;

s2: identifying the collected data through an RBF classifier;

s3: and fusing the identification results of the RBF classifier by using a D-S evidence theory to obtain a final identification result.

2. The fusion recognition method for the coal mining machine coal rock cutting state according to claim 1, characterized in that the S1 process respectively comprises a sound sensor and two vibration sensors.

3. The fusion recognition method for the coal cutter coal rock cutting state according to the claim 1 or 2, characterized in that in the S2 process, the RBF classifier comprises an input layer, a hidden layer and an output layer, wherein the function of the RBF is expressed as:

wherein, | | x_p-c_i||²Is the norm of the euclidean norm,

is the p-th input sample, c_iIs the center of the gaussian function of the ith node, and σ is the variance, i.e., the width of the gaussian function;

the output calculation method of the RBF classifier comprises the following steps:

wherein ω is_ijIs the connection weight between the hidden layer and the output layer, h is the number of nodes in the hidden layer, θ_jIs the threshold for the jth output node, which is the actual output value of j-, the first output node in the output layer.

4. The fusion recognition method for the coal cutter coal rock cutting state according to claim 1 or 2, wherein in the S2 process, an MFO-FOA hybrid optimization algorithm is adopted in an RBF classifier for classification, a random flight pattern of the fruit flies is changed into a spiral flight path, the center of the spiral flight path is the current position of the fruit fly population, and the specific algorithm is as follows:

where dis _ X_iAnd dis _ Y_iRespectively, a single position X of the previous generation_{i_last}And Y_{i_last}With current group position (X)_{i_last}And Y_{i_last}) X _ axis and Y _ axis are the initial positions of the drosophila population;

the specific implementation steps are as follows:

t1: setting a population size N, a maximum iteration number T, a position range and a random close range FR of a fruit fly population LR; randomly generating an initial Drosophila population position (X _ axis, Y _ axis); the position of the individual fruit flies is then updated:

t2: calculating the distance between the current position and the origin to obtain an odor concentration judgment value (S)_i)

T3: judging the odor concentration (S)_i) Input to (Smell)_i) In the judgment function, calculating the odor concentration Smelli of the y position of a single fruit; the best odor concentration was obtained by ranking all Smelli, and then determining and finding the corresponding single position bestIndex:

t4: comparing the current bestmdex with the global optimum concentration Smellbest; if bestIndex outperforms Smellbest, the global optimum concentration and corresponding location are updated:

t5: individual positions of fruit flies were updated:

t6: if the termination condition is met, stopping iteration; otherwise, steps T2-T5 are repeated until the maximum number of iterations T is reached.

5. The fusion recognition method for the coal-mining machine coal-rock cutting state according to claim 1 or 2, characterized in that in the S3 process, evidence correlation coefficient measurement is introduced, a weight-mean combination model is used to perform coefficient operation between the evidences and modify or preprocess the original evidences;

for two evidences m in the recognition framework Θ_iAnd m_jThe correlation coefficient r is calculated by the following formula_BPAAnd degree of correlation c:

m is to be_iAnd m_jSupport degree of correlation coefficient between them as evidence, Sup (m)_i，m_i)＝r_BPA(m_i，m_j) Simply referred to as S_ijObtaining a support matrix of the evidence:

other evidence is evidence m_iThe total support of (1) is:

the normalized support can be used as evidence m_iReliability of (c) Crd (m)_i) The calculation formula is as follows:

confidence level Crd (m)_i) As a weight given to the evidence, a new evidence is obtained by averaging all evidences according to the weight, achieving fusion of highly conflicting evidences.

Images